TensorsOfTheWallIndustry: Where I learned that "It works on my machine" is not an acceptable debug strategy.
Click on each card to view more details.
Summer Intern|Full-Time|Hyderabad, India
Achievements
- Developed proof-of-concept for upgrading existing product on ARM hardware using embedded C++, delivering nearly 2x processing power while reducing manufacturing and development costs by 70%
- Designed user interface with Qt5 and implemented on Linux Yocto, enabling custom Linux-based systems independent of hardware architecture for seamless user interaction
Software Engineer|Full-Time|Bangalore, India
Chose to pursue graduate studies after successfully completing a 1-year fixed-term role at Synopsys, where a full-time offer was extended
Achievements
- Optimized existing data extraction and processing pipelines with Apache Spark, improving processing time by 25%
- Developed BERT-based anomaly detection tool to identify, track, and alert users to critical events requiring intervention, achieving average accuracy of 92% in runtime issue detection
- Redesigned several existing applications for log analysis into a single unified system in production, improving performance by 30% while enhancing user experience, reliability, and accessibility
Software Engineer (AI & Machine Learning)|Full-Time [Contract]|Bangalore, India
Achievements
- Improved inference performance on Untether AI's accelerator chips by developing high-performance low-level INT8 kernels in C++
- Implemented generalized padding for transformation kernels, replacing the non-generalized version, and optimized quantization, performance metrics, and execution cycles, achieving 30% performance gain
- Led team to streamline workflows, create detailed documentation, and reduce codebase size by 40%
Software Engineering Intern|Full-Time|Waltham, MA
Achievements
- Developed data summarization algorithms for the Ottoviz platform by creating a custom fork of the submodlib library and applying submodular optimization techniques.
- Enabled efficient training of deep learning models by developing custom features for selecting important 1% subset of images, achieving significant dataset reduction.